AI visibility
Citation, mention or retrieved source: how to tell them apart
A citation puts the brand in the generated text, a mention names it without building on it, and a retrieved source is what the engine read. Mixing the three makes visibility look bigger than it is.
A citation is a direct, attributable reference to the brand. A mention names the brand without building any real point on it. A retrieved source is content the engine pulled in and used to compose its answer. The three are worth separating because each one tells you something different: a citation is about presence in the generated text, a mention is about recognition, and a retrieved source is about what the engine actually read. When you look at brand citations on ChatGPT, or on any other engine, blending these levels means reading a passing phrase as if it were visibility.
Why three categories instead of one
People who watch generative engine answers tend to call everything a citation. It is a convenient habit, and an imprecise one. A brand can show up in an answer because the model named it on its own, because it found the brand on a retrieved page, or because that page cited the brand explicitly. Those are three different paths and they call for three different readings. The classification below is a working proposal from RankAI. It is not a shared standard or an industry convention: we offer it as a way to read the evidence, to be used and corrected.
The three categories, one at a time
Citation
We call it a citation when the brand is named directly and the reference can be attributed: the name appears, the context identifies it, and a reader understands that this is the brand in question and not another company with the same name. It is the strongest of the three, because it still holds up when the text is read outside the context of the answer.
Mention
A mention is weaker. The brand is named, but the passage does not build any useful information on it. It might be a list, a parenthetical, an example. It is still a signal, but it should not be counted as a citation. Treating it as one inflates the sense of presence and makes it harder to see which answers really matter.
Retrieved source
Here the reading model is ours, and it should be taken as such. We call a retrieved source any content the engine pulled in to compose the answer, whether or not the brand appears in it as a citation or a mention. This is not a verified product fact. It is a way of separating what the engine read from what it wrote, so that the brand's presence in the answer is not confused with the brand's presence on the pages the engine used.
Telling them apart in practice
The test is the attribution chain. If the brand is in the generated text and can be identified, it is a citation. If it is in the text but does not build a point, it is a mention. If the brand is not in the text but content about it was retrieved, you are at the third level. One practical detail: when any of these measurements is not available, RankAI shows it as N/A and never as zero, because no measurement is not the same as a measurement of absence. That applies to all three categories and keeps a missing figure from being read as a negative one.
Original contribution from RankAI. The citation / mention / retrieved source split is a working proposal from RankAI, not a universal standard. Our definition of "retrieved source" is a reading model we offer to separate what the engine read from what it wrote. It is not a verified product fact and should not be treated as a given category. Anyone who adopts this scheme should test it on their own cases and adapt it.
Limits
This classification says nothing about the quality of an answer or about a brand's ranking. It only describes how to read the three kinds of evidence. RankAI's public pages name ChatGPT, Perplexity, Gemini, Grok and Google AI. Claude and Grok are measured only after you connect them, and Gemini and Google AI Overviews are not the same observation. For the full picture of how we read visibility, see the in-depth page.